Why It's So Hard to Count Twitter Bots

Is the account real? It's difficult to quantify the proportion of fake accounts across the social network because one of the best methods for detecting fake accounts thinks it might be.

There is a point of contention in the ongoing $44 billion acquisition of Twitter. The billionaire put his purchase on hold until the company provided details to back up its claim that less than 5 percent of its shares were owned by foreigners. A plan to count bots was outlined by Musk, which involved sampling 100 followers to see how many were fake.

It's more difficult to accurately quantify the percentage of bots on the micro-messaging service.

If you know where to look, you can find them. Chris Bail, a professor of sociology, says that certain accounts, including Musk's, seem to attract a lot of them.

There are other social networks that struggle with fake accounts. Billions of fake accounts are removed from Facebook every year. It is difficult to know if an account is a bot since legitimate users may not have many followers, or have strange usernames. It's difficult to gauge the number of bots on the platform as a whole.

V.ai, an artificial intelligence company that has previously identified bot-like activity among accounts that amplify misinformation about US voter fraud, looked at 100 accounts that follow Musk's car manufacturing company.

Out of 100 accounts, more than 20 have a high likelihood of being bots. More than half of the 100 were found to be bots. There was no evidence that any of the accounts were promotional. Many of the accounts disappeared shortly after, suggesting that the social networking site catches bots quickly. Identifying dubious accounts is inherently subjective and involves a degree of uncertainty, says Vince Lynch, CEO of IV.ai.

Filippo Menczer, a professor at Indiana University who led the development of the Botometer algorithm, says that it is a very hard problem. Menczer says that looking at 100 accounts won't tell the whole story, and different samples will produce wildly different results.